Background
Over the past few years, the path for both Alzheimer’s disease (AD) research and diagnosis has been radically changed due to developments in the field of biomarkers, as highlighted in the recent National Institute on Aging and Alzheimer’s Association biological definition of AD [
1]. Different modalities of AD biomarkers have been implemented, including both neuroimaging and cerebrospinal fluid (CSF) biomarkers. In CSF, a combination of low levels of the 42-aminoacid isoform of amyloid beta (Aβ42) and high levels of total tau (t-Tau) and phosphorylated tau (p-Tau) is thought to reflect the two widely accepted pathophysiological hallmarks of AD: amyloid plaques and neurofibrillary tangles [
2]. In clinical practice, these biomarkers are useful to detect or exclude AD, to make a prognosis at the Mild Cognitive Impairment (MCI) stage, and to guide patients’ management, particularly in atypical and clinically challenging cases [
3,
4]. These biomarkers have also been incorporated in clinical trials, not only for patient selection—in fact, it was found in past AD drug trials that many individuals enrolled did not have AD brain pathology—but also to monitor target engagement and eventually as surrogate end points [
5]. When an effective drug for AD is available, CSF biomarkers will become even more important in guiding the diagnosis and management of clinical cases.
However, the use of CSF biomarkers as diagnostic devices worldwide is hampered by problems of comparability of the results obtained in different centers or on different analytical platforms, low specificity towards non-AD cognitive diseases at the MCI stage, and limited understanding on how to interpret results, particularly if they seem discordant versus other biomarker modalities [
6]. Until now, the INNOTEST enzyme-linked immunosorbent assays (ELISA) have been the mostly used assays for routine CSF biomarker analysis. These assays involve several manual pipetting steps, resulting in over 15% inter-laboratory variation of results, as reported in the Alzheimer’s Association international quality control program (
www.neurochem.gu.se/TheAlzAssQCprogram) [
7]. Another problem of these assays is the quite long turnaround time, as usually laboratories tend to accumulate samples over time, until they have enough to fill in an ELISA 96-well plate. Moreover, for the INNOTEST, some authors have reported an upward drift in Aβ42 values over time [
8,
9].
Several international standardization initiatives have been launched to improve intra- and inter-laboratory variability, by standardizing pre-analytical variables, analytical protocols. and assay calibrators [
10‐
12]. Although major advances have been made in the field [
13], the situation is still not optimal, and universally accepted cut-offs have not been reached. To reduce variation in manual immunoassays and to cope with the increase in the number of referrals, multiplex assays and (semi) automated platforms have been developed [
14‐
17]. Recently, four CSF analytes (Aβ42, Aβ40, t-Tau, and p-Tau) have been implemented on the fully automated Lumipulse G System, which is based on Chemiluminescent Enzyme Immunoassay technology. Lumipulse G uses single-analyte, ready-to-use, immunoreaction cartridges and renders quantitative results for an analyte within 30 or 35 min on the LUMIPULSE G1200 and G600II, respectively. These assays typically show an inter-laboratory variability of less than 10% (
www.neurochem.gu.se/TheAlzAssQCprogram), but data regarding their clinical validation in research cohorts is still very limited [
18‐
20]. Cut-offs that optimize the agreement between CSF biomarkers measured on the LUMIPULSE G600II instrument and amyloid imaging results by 18F-Florbetapir PET have been reported [
21], but there are no validated cut-offs for these four CSF biomarkers in relation to clinical AD diagnosis.
The aims of this study were to (i) evaluate the analytical performance of the Lumipulse G β-Amyloid 1-42, β-Amyloid 1-40, total Tau, and pTau 181 assays on the fully automated LUMIPULSE G600II platform; (ii) compare CSF biomarker results of the Lumipulse G assays with the established manual ELISA assays (INNOTEST® β-AMYLOID(1-42), INNOTEST β-AMYLOID(1-40), INNOTEST hTAU Ag, and INNOTEST PHOSPHO-TAU(181P)); and (iii) establish cut-offs and the clinical performance of the Lumipulse G assays for AD diagnosis.
Discussion
Our results show that the Lumipulse G β-Amyloid 1-42, β-Amyloid 1-40, total Tau, and pTau 181 assays on the fully automated LUMIPULSE G600II platform have a very good analytical performance. In our hands, the inter-assay coefficients of variation ranged between 0.66 and 3.25%, while the intra-assay coefficients of variation varied between 0.79 and 5.50%. These values are in line with what was recently reported by Bayart and colleagues [
20], are within what is desired for a routine diagnostic test, and are lower than what has been reported for INNOTEST and other ELISA assays, both by the manufacturer and by independent studies [
45]. In addition to these analytical characteristics, Lumipulse assays also showed an excellent diagnostic accuracy for AD, reaching sensitivity and specificity levels from around 80% (in the case of Aβ42 alone) to up to more than 95% (for ratios between markers). These figures are at least as good as the ones generally reported for ELISA assays [
46].
One of the main goals of this work was to establish cut-offs for the CSF biomarkers and their ratios, analyzed using the Lumipulse G platform, for the clinical diagnosis of AD. To the best of our knowledge, no other study has reported such cut-offs for all four biomarkers. The study by Alcolea and colleagues [
21] included 94 participants from the Sant Pau Initiative on Neurodegeneration (SPIN cohort), but determined cut-offs for the Lumipulse assays by optimizing their agreement with 18F-Florbetapir PET amyloid imaging results, and not to the clinical diagnosis. Moreover, the population used was much more heterogeneous, including non-AD dementia cases. Therefore, the reported cut-offs of the three markers were quite different from ours. Interestingly, however, the cut-offs for the Aβ42/Aβ40 and Aβ42/t-Tau ratio were quite similar. The work of Paciotti and colleagues [
19] compared AD (
n = 42) and non-AD (
n = 38) patients, assessing the diagnostic accuracy of only Aβ42 and t-Tau Lumipulse assay to distinguish between the two groups, but did not report the cut-off values. The recent work of Bayart and co-workers [
20] used 44 AD patients and 42 controls to establish cut-offs for Lumipulse Aβ42 and t-Tau, but not for p-Tau or the Aβ42/40 ratio. These authors reached values of 437 pg/mL for Aβ42and 381 pg/mL for t-Tau, slightly different from ours, particularly for Aβ42. Apart from this small study, the only cut-offs for these assays that we are aware of and that were established based on clinical diagnosis are the ones recommended by the manufacturer. These were calculated based on the comparison of 60 probable AD patients and 43 non-demented controls (other neurological disorders such as psychiatric disorders, epilepsy, and multiple sclerosis), using a statistical approach similar to ours (ROC curve analysis with cut-offs selected based on maximal Youden index). Although slightly higher, the cut-offs for Aβ42, t-Tau, and p-Tau are not very different from ours (599 pg/mL, 404 pg/mL, and 56.5 pg/mL, respectively). The small differences between our cut-offs and previously reported ones could be attributed to the characteristics of the population or deviations in the pre-analytic protocol. In our control population, similarly to what is reported by Bayart et al. [
20], we included cognitively normal patients with a suspicion of a neurological disease, but in whom a major CNS disease was excluded. However, while our control group includes mainly idiopathic headaches and some peripheral polyneuropathies, the control population that was selected by Bayart and colleagues is much more heterogeneous, including a large diversity of diagnosis. In relation to the control population used by the manufacturer, other non-neurodegenerative neurological diseases were included, and that could account for the differences in t-Tau and p-Tau cut-offs. Moreover, as shown in Table
2, our population is quite young, probably due to the fact that it comes from a specialized memory clinic, and that could also add to the differences in t-Tau and p-Tau levels [
47]. The fact that we noticed the same trend for our INNOTEST cut-offs also argues for it being related to the population or pre-analytical confounders rather than the assays. Although the pre-analytical protocol that we used was similar to the one used by the manufacturer’s and by Bayart et al., there are slight differences, particularly in relation to the study of Bayart and colleagues, in relation to the type of tubes used for aliquoting, and filling of the tubes, that could justify the small variation in the cut-offs, particularly for Aβ42 [
10,
48]. Noteworthy, our cut-off for the Aβ42/Aβ40 ratio is practically the same as the one recommended in the package insert (0.069), reinforcing the notion that this ratio is a more robust and easily standardized marker.
Although our study was limited by the relatively small sample size, a few points make us confident in the established cut-offs. First, when we re-calculated our cut-offs using only the subset of AD patients that had a confirmatory amyloid PET imaging result, the values reached were essentially the same as for the whole cohort. Second, the diagnostic accuracy derived from these cut-offs of both the Lumipulse assays and their ratios was at least as good as the one we have previously reported for the INNOTEST assays, employing larger cohorts of AD and neurological controls but with similar characteristic than the one included here [
39,
43]. Moreover, we performed a small validation of our cut-offs in an independent cohort, which showed a good accuracy, correctly classifying 83% (for Aβ42 alone) to 97% (for ratios between markers) of the individuals. Interestingly, in this validation cohort, all controls were correctly classified by all markers and all three ratios performed exactly the same. If we compare the accuracy figures of this validation cohort (Table
3) with the values depicted in Fig.
2c for the discovery cohort, the total diagnostic accuracy is similar, although the data for the validation cohort seems in favor of the specificity. However, this validation population is very small, and further studies are needed to fully evaluate the accuracy of this cut-offs, ideally with a multi-center design.
As recently shown by others [
20,
21,
49], a strong correlation between Lumipulse and INNOTEST Aβ42 and t-Tau assays was observed. Aβ40 and p-Tau also showed good correlations, in line with the work of others [
20,
50,
51]; however, both a systematic and a proportional difference between methods was observed. Although both systems use similar antibody combinations, the discrepancy between absolute levels measured by both platforms might be due to differences in the technology. Furthermore, for Aβ40, the need to dilute samples in the ELISAs that induce an extra source of error might explain the slightly lower correlation coefficient for this assay. For p-Tau, although the correlation is good, in our cohort, Lumipulse p-Tau values seem to be higher than INNOTEST in the high range, and a bit lower in the low range. Nevertheless, the discrimination between AD and controls was better for the Lumipulse than for the INNOTEST.
In spite of differences in the absolute values for the various markers and ratios, both methods classified individuals in a similar way, with overall percentages of agreement of classification between 87.5 and 97.5%. Interestingly, in the majority of cases with discordant results in at least one of the biomarkers or their ratios (23 out of 29), Lumipulse classification was in agreement with the clinical diagnosis. Concordance analyses of biomarker abnormality based on cut-points are relevant to allow method comparisons on an individual level. When applying the NIA-AA criteria [
1] to classify subject as having or not a CSF-AD profile, we observed that 12 cases (10%) were discordant according to the analytical method used. However, in all of these discordant cases, the marker that was discordant (either p-Tau or the Aβ42/Aβ40 ratio) had values near the cut-off that were within the usually called border zone [
44]. Biomarker values near the cut-point need to be interpreted with caution, as technical or biological variation can influence the absolute values. Therefore, results within this border zone should be interpreted as at risk for abnormality rather than a conclusive positive or negative outcome.
One of the findings of our study was that the combination of Aβ42 with a second marker, either another amyloid marker (Aβ40) or a neurodegeneration/fibrillary tau marker (t-Tau or p-Tau), resulted in significant increases of accuracy for all cases, with the three ratios reaching a similar diagnostic accuracy. Therefore, our results confirm the superior value of the ratios and also highlight the use of the Aβ42/Aβ40 to compensate for individual differences in amyloid precursor protein processing that otherwise would result in an incorrect interpretation of Aβ42 CSF results [
52]. Moreover, it has been shown that the CSF Aβ42/Aβ40 ratio can better predict abnormal cortical amyloid deposition compared with CSF Aβ42 [
53,
54] and compensate for the effects of pre-analytical interfering factors, such as tube type, freeze/thaw cycles, and CSF volumes, therefore contributing towards pre-analytical standardization [
55,
56]. Worth mentioning, in our analysis in the subgroup of 35 AD patients with positive amyloid imaging, three had a normal Lumipulse Aβ42 result, while the Aβ42/Aβ40 ratio was abnormal in all cases. Our results therefore support the use of the Aβ42/Aβ40 ratio in clinical care settings.
We believe that one of the main strengths of our study relies in the study design: the four AD CSF biomarkers (Aβ42, Aβ40, t-Tau, and p-Tau) were measured simultaneously, from the same aliquot, using both the Lumipulse and INNOTEST assays; the same batch of reagents for each marker/assay was used and a standard CSF pre-analytical procedure was followed throughout the study. In addition, the Lumipulse Aβ42 levels were standardized to the recently developed CRM, therefore allowing comparison with future studies. However, some limitations of the current study must also be addressed. In our study, some samples had been stored for quite a long time and this might have influenced the absolute levels of the different markers measured. However, a previous study has showed stable CSF Aβ42, t-Tau, and p-Tau concentrations over 12 years of biobank storage [
57]. As participants in this study are part of a living cohort, neuropathological confirmation was not available, leaving the possibility of misdiagnosis. We tried to circumvent this problem by including only patients with a clinical AD diagnosis with a high degree of certainty, either due to a confirmatory amyloid imaging test or to their long clinical follow-up. A major limitation of this study is the small sample size. As explained above, by including only AD patients with a high degree of certainly, we limited the number of patients that could be included in the analysis. Moreover, as our population comes from a specialized memory clinic, having assess to neurological control samples is also a major difficulty. To address this extremely important issue of sample size, we propose that a multi-center study, including a large number of subjects and involving different laboratories that already have experience with both assays (Lumipulse and INNOTEST), should be conducted.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.